I've seen fantastic research sunk and forced through resubmission time and again because of reviewers that either don't understand the work (and why would they? they have zero incentive) or have pre-existing biases based on their own work (and are called in as expert reviewers, effectively functioning as gatekeepers in a subfield).

Bad deans/department chairs can make life suck pretty hard. This is usually not the case because they (in my experience) don't have that much power, but over some things like hiring they can definitely be as capricious as they want. The real killer are the people outside the department, on IRBs, grant review boards, etc.. In a good research-focused school they can be great; in worse schools they can be tinpot despots.

Reproducibility in particular is insane. In computer science virtually no systems are released at the time of a paper submission, so papers describe things that may or may not exist. I'd say this is less due to untrained/inexperienced grad students and more due to a combination of constantly rushing to publish, unclear institutional regulations regarding releasing artifacts such as source, and inability to focus on cleaning things up for a release.

Academia has serious problems at its core, and in a lot of places, it's obvious that startups and industry is outpacing it. I think this comes down to an incentive problem -- when there's no incentive to build things that work, you tend to build broken things.